- Email: [email protected]
Taste sensor
Sensors and Actuators B 64 Ž2000. 205–215 www.elsevier.nlrlocatersensorb
Taste sensor Kiyoshi Toko Department of Electronic DeÕice Engineering, Graduate School of Information Science and Electrical Engineering, Kyushu UniÕersity, Fukuoka, 812-8581, Japan
Abstract A taste sensor with global selectivity, i.e., electronic tongue, is composed of several kinds of lipidrpolymer membranes for transforming information of taste substances into electric signal. The sensor output shows different patterns for chemical substances which have different taste qualities such as saltiness and sourness. Taste interactions such as suppression effect, which occurs between bitterness and sweetness, can be detected and quantified using the taste sensor. Amino acids and peptides can be classified into several groups according to their own tastes from sensor outputs. Bitter-tasting amino acids such as L-tryptophan have response electric patterns similar to a typical bitter substance, quinine. The taste of foodstuffs such as beer, sake, coffee, mineral water, milk and vegetables can be discussed quantitatively. The taste sensor with lipid membranes provides the objective scale for the human sensory expression and will contribute to clarification of the reception mechanism at gustatory cells. q 2000 Elsevier Science S.A. All rights reserved. Keywords: Taste sensor; Chemical substances; Global selectivity; Quantified taste; Amino acids; Lipid membranes
1. Introduction Taste is comprised of five basic qualities w1,2x: sourness produced by hydrogen ions of HCl, acetic acid, citric acid, etc.; saltiness produced mainly by NaCl; sweetness due to sucrose, glucose, etc.; bitterness produced by quinine, caffeine and MgCl 2 . The last is umami taste, which is the Japanese term for implying ‘‘deliciousness,’’ produced by monosodium glutamate ŽMSG. contained in seaweeds, disodium inosinate ŽIMP. in meat and fish and disodium guanylate ŽGMP. in mushrooms. In a gustatory system, substances producing taste are received by the biological membrane of gustatory cells in taste buds on tongue. Information on taste substances is transduced into an electric signal, which is transmitted along the nerve fiber to the brain, where the taste is perceived. We have developed a taste-sensing system whose transducer is composed of lipidrpolymer membranes w3–5x. The output of this system is not the amount of specific taste substances but the taste quality and intensity, because different output electric patterns are obtained for chemical substances producing different taste qualities such as sourness and saltiness. On the other hand, similar patterns are obtained for chemical substances producing the same taste, such as MSG, IMP and GMP, which have an umami taste, and NaCl, KCl and KBr for saltiness. The development of this sensor is based on a concept very different from that of conventional chemical sensors,
which selectively detect specific chemical substances such as glucose or urea. However, taste cannot be measured even if all the chemical substances contained in foodstuffs are measured. Humans do not distinguish each chemical substance, but express the taste itself; the relationship between chemical substances and taste is not clear. It is also not practical to arrange so many chemical sensors as chemical substances, which number over 1000 in one kind of foodstuff. Moreover, there exist interactions between taste substances, such as a suppression effect. Sweet substances suppress the taste intensity of bitter substances. Discrimination of each chemical substance is not important here, but recognition of the taste itself and its quantitative expression must be made. The taste sensor using lipidrpolymer membranes has a concept of global selectivity, which implies the ability to classify enormous kinds of chemical substances into several groups.
2. Taste sensor 2.1. Multichannel electrode Eight kinds of lipids were used for preparing the membranes; used lipids were, e.g., oleic acid ŽOA., oleyl amine ŽOAm. and decyl alcohol ŽDA. as listed in Table 1. Depending upon the object to be measured, we prepared
0925-4005r00r$ - see front matter q 2000 Elsevier Science S.A. All rights reserved. PII: S 0 9 2 5 - 4 0 0 5 Ž 9 9 . 0 0 5 0 8 - 0
K. Toko r Sensors and Actuators B 64 (2000) 205–215
206
Table 1 Lipid materials used in the multichannel electrode Channel
Lipid Žabbreviation.
1 2 3
n-Decyl alcohol ŽDA. Oleic acid ŽOA. Dioctyl phosphate ŽBisŽ2-ethylhexyl.hydrogen phosphate, DOP. DOP: TOMA s9:1 DOP: TOMA s 5:5 DOP: TOMA s 3:7 Trioctyl methyl ammonium chloride ŽTOMA. Oleyl amine ŽOAm.
4 5 6 7 8
different lipid materials. For example, the mixed, hybrid membranes composed of dioctyl phosphate ŽDOP. and trioctyl methyl ammonium chloride ŽTOMA. were used for measurements of amino acids w6x. The lipidrpolymer membrane was a transparent, soft film with approximately 200 mm thickness. Each lipidrpolymer membrane was fitted on the part of a plastic tube, which has a hole, such that the inner part of the cylinder is isolated from the outside. The end of the cylinder was sealed with a stopper that holds an AgrAgCl electrode. The tube was filled with 3 M KCl solution. Eight detecting electrodes thus prepared were separated into two groups, and connected to two electrode holders, which are controlled mechanically by a robot arm, as illustrated in Fig. 1. 2.2. Responses to fiÕe primary tastes Fig. 2 shows the electric potential pattern from seven channels for two kinds of taste qualities of sourness and
saltiness, measured by taking the origin to 1 mM KCl w7x. The patterns of substances producing different taste qualities are much different, and hence, each taste can be easily discriminated. The reproducibility was very high, because the standard deviations were smaller than 1%. On the other hand, the taste sensor has similar response patterns to the same group of taste, i.e., as examples of sour substances, HCl, citric acid and acetic acid show similar response patterns. Salty substances, NaCl, KCl and KBr show similar patterns, too. The same result also holds for other taste qualities such as bitterness, sweetness and umami taste. The taste sensor can respond to the taste itself. This fact implies that the taste sensor has global selectivity. Of course, the taste sensor has an ability of molecular recognition to distinguish chemical substances even in the same group of taste, e.g., the patterns for HCl, citric acid and acetic acid are slightly different. The reception mechanism in lipidrpolymer membranes of the taste sensor was clarified quantitatively based on electrochemical theory, which treats the surface electric potential, surface charge density and binding of ions such as protons and hydrophobic ions w8,9x. This result suggests participation of a similar mechanism at the gustatory reception level.
3. Amino acids Each channel responded to amino acids in different ways depending on their tastes w6x. L-Tryptophan, which elicits almost pure bitter taste in humans, increased the potentials of channels 1, 2 and 3 greatly. This tendency was observed for other amino acids, which mainly exhibit
Fig. 1. Taste-sensing system ŽSA402, Anritsu..
K. Toko r Sensors and Actuators B 64 (2000) 205–215
207
Fig. 2. Response electric potential patterns for sour and salty substances w7x.
bitter taste, such as L-phenylalanine and L-isoleucine. On the other hand, L-alanine, glycine and L-threonine taste mainly sweet. For these amino acids, the potentials of channels 1 and 2 decreased. L-Valine and L-methionine, which taste bitter and slightly sweet, decreased the potential of channel 5; the responses of channels 1 and 2 were small. Fig. 3 shows a diagram obtained by the principal component analysis ŽPCA. w6x. The first principal axis ŽPC1. reflects bitterness and sweetness, while the second principal axis ŽPC2. reflects sourness and umami taste. Amino acids are roughly classified into five groups. The taste changes gradually from sweetness for glycine to bitterness for L-tryptophan along the PC1 axis. In addition, a strong correlation between the output of the taste sensor and hydrophobicity of amino acids was found w10x. Next, we compared the response patterns for bitter-tasting amino acids such as L-tryptophan with that of quinine, which is a typical bitter substance. Of course, these two chemicals have very different chemical structures; nevertheless, we feel the same bitter taste quality. Fig. 4 shows the comparison of response patterns of three amino acids ŽL-alanine, L-tryptophan, L-phenylalanine. with those of a bitter substance Žquinine., sour substance ŽHCl. and umami taste substance ŽMSG. w11x. The response patterns were normalized using the formula: Õi s
Vi
(
,
8
Ý < Vi <
Fig. 4 includes three normalized patterns for one chemical substance with three different concentrations. By the normalization procedure of Eq. 1, the three patterns of each chemical substance agree with each other; that is, the pattern is independent of the concentration. This fact implies that each chemical substance has an original pattern characteristic of each taste quality. The patterns of amino acids such as L-tryptophan and L-phenylalanine are similar to that of quinine. However, they differ from those of other taste substances such as L-alanine, HCl and MSG. The patterns for L-tryptophan and L-phenylalanine jut out to the upper-right direction Ži.e., at channels 1–3., whereas the pattern for L-alanine shows the opposite tendency: its pattern bulges out to the lower-left direction. The bitter chemical quinine shows a bulge to the upper-right direction in a similar way to L-tryptophan.
Ž 1.
2
is1
where Vi denotes the response electric potential of channel i. To make it easy to see the pattern, the response electric potentials of the positively charged membranes Žchannels 6–8. in Table 1 and the non-charged membrane Žchannel 5. were reversed, because they usually have a sign opposite to those of the negatively charged membranes Žchannels 1–4..
Fig. 3. Discrimination of taste of amino acids using the taste sensor w6x.
208
K. Toko r Sensors and Actuators B 64 (2000) 205–215
Fig. 4. Normalized patterns for L-tryptophan Ža., L-phenylalanine Žb., quinine Žc., L-alanine Žd., HCl Že., and MSG Žf. w11x. The three numerical figures closed by a square attached to each pattern imply the concentration ŽmM..
The correlation coefficients of the pattern between Ltryptophan Ž10 mM. and five basic taste substances are as follows: 0.903 for 0.3 mM quinine, 0.276 for 30 mM NaCl, 0.763 for 3 mM HCl, 0.515 for 100 mM sucrose and 0.408 for 10 mM MSG. Although L-tryptophan has low correlations with salty ŽNaCl., sweet Žsucrose. and umami ŽMSG. taste substances, it has a high correlation with a bitter substance, quinine. This result indicates that L-tryptophan shows the same taste as quinine, i.e., L-tryptophan tastes bitter.
Comparison of the original response pattern of L-tryptophan with that of quinine makes it possible to estimate the bitter strength of L-tryptophan in terms of the quinine concentration Žsee also Fig. 9.. As a result, it was concluded that 10 mM L-tryptophan has the same bitter strength as 0.02 mM quinine. To confirm it, we performed the sensory evaluation using our tongue. The result supported the estimation using the taste sensor: humans felt the same bitter strength as 10 mM L-tryptophan by taking 0.02–0.03 mM quinine.
K. Toko r Sensors and Actuators B 64 (2000) 205–215
Fig. 5. Normalized patterns for L-methionine Ža. and the mixed solution Žb., which contains 100 mM L-alanine and L-tryptophan with three different concentrations Ž1, 3, 10 mM. w11x.
This result implies that the taste sensor measures the taste in itself, as humans do, irrespective of the difference of chemical structures of amino acids and alkaloids such as quinine. L-Methionine and L-valine shows bitter and sweet tastes simultaneously w12–14x. We tried to produce the complicated mixed taste of these amino acids using the combina-
209
tion of a sweet amino acid, L-alanine, and a bitter amino acid, L-tryptophan. Fig. 5 shows the normalized patterns for L-methionine and the mixed solution, which contains 100 mM L-alanine and L-tryptophan at 1, 3 and 10 mM w11x. Three patterns for L-methionine at three different concentrations agree fairly well with each other. In addition, they differ from those for either L-alanine or L-tryptophan alone, as shown in Fig. 4. For example, 10 mM L-methionine has the low correlations of 0.22 and 0.57 with 10 mM L-alanine and 1 mM L-tryptophan, respectively. This implies that Lmethionine has an original taste quality that is different from pure sweet or bitter taste. The patterns for the mixed solution composed of 100 mM L-alanine and L-tryptophan of three different concentrations do not agree with each other. Increasing the Ltryptophan concentration tends to cause a bulge to the upper-right direction. Since the pattern for L-tryptophan alone has a characteristic bulge of this type, it implies that the taste becomes nearer that of L-tryptophan, as expected. Table 2 summarizes the correlation coefficients among L-methionine, L-alanine, L-tryptophan and the above three mixed solutions w15x. The response patterns for L-alanine have negative correlations with those for L-tryptophan. It agrees with the fact that they show very different taste qualities. We can see that the highest correlation is realized between one of the mixed solutions and L-methionine. Increasing L-tryptophan concentration in the mixed solution leads to both the decrease of correlation with L-alanine and the increase of correlation with L-tryptophan, as considered reasonably. The highest correlation 0.99 is realized for 30 mM L-methionine using the mixed solution 100 mM L-alanine and 10 mM L-tryptophan. The sensory tests by humans agreed with this result. A similar result was obtained for L-valine, which elicits bitter and sweet tastes simultaneously. The sensor output showed the highest correlation between 30 mM L-valine and 100 mM L-alanine plus 3 mM L-tryptophan, as felt by humans.
Table 2 Correlation coefficients among L-methionine, L-alanine, L-tryptophan and the three mixed solutions w15x. The unit of the concentrations is mM; e.g., for L-methionine, 10, 30, and 100 mM were chosen. In the case of mixed solutions composed of 100 mM L-alanine and L-tryptophan, the L-tryptophan concentration ŽmM. is shown Methionine 10 Methionine
Mixed
Alanine
Tryptophan
10 30 100 1 3 10 10 30 100 1 3 10
0.96 0.93 0.93 0.96 0.93 0.22 0.26 0.34 0.57 0.60 0.61
Mixed
Alanine
30
100
1
3
10
0.96
0.93 0.99
0.93 0.90 0.89
0.96 0.97 0.96 0.95
0.93 0.99 0.98 0.88 0.98
0.99 0.90 0.97 0.99 0.10 0.13 0.23 0.66 0.70 0.73
0.89 0.96 0.98 0.11 0.14 0.24 0.64 0.68 0.72
0.95 0.88 0.48 0.52 0.60 0.33 0.38 0.40
0.98 0.20 0.24 0.34 0.59 0.64 0.65
0.02 0.06 0.16 0.72 0.76 0.78
10 0.22 0.10 0.11 0.48 0.20 0.02 1.00 0.99 y0.66 y0.62 y0.61
Tryptophan 30 0.26 0.13 0.14 0.52 0.24 0.06 1.00 0.99 y0.63 y0.59 y0.58
100 0.34 0.23 0.24 0.60 0.34 0.16 0.99 0.99 y0.56 y0.51 y0.49
1
3
10
0.57 0.66 0.64 0.33 0.59 0.72 y0.66 y0.63 y0.56
0.60 0.70 0.68 0.38 0.64 0.76 y0.62 y0.59 y0.51 1.00
0.61 0.73 0.72 0.40 0.65 0.78 y0.61 y0.58 y0.49 0.99 1.00
1.00 0.99
1.00
210
K. Toko r Sensors and Actuators B 64 (2000) 205–215
Dipeptides elicit various taste qualities according to the combination of two amino acids. For bitter dipeptides such as Gly–Leu, Gly–Phe and Leu–Gly, the taste sensor showed similar response patterns for L-tryptophan. In addition, the patterns characteristic of sourness, which is elicited by amino acids such as L-glutamic acid and L-histidine monohydrochloride, were obtained for dipeptides, such as Gly–Asp, Ser–Glu, Ala–Glu and Gly–Gly, which taste sour. On the other hand, dipeptides such as Ala–Gly and Gly–Glu have little or no taste. Small response patterns were obtained for these dipeptides, as expected. It is believed that receptors of amino acids differ physiologically from those of alkaloids such as quinine. As found here, however, the bitterness of amino acids can be expressed using the response patterns of the taste sensor, which uses the lipid membranes, in a way similar to the bitterness of quinine. The mixed taste of amino acids can also be produced using the taste sensor. This result suggests that it may be the lipid Žhydrophobic. part of biological membranes that forms the receptor for bitter taste. 4. Quantification of taste of foods and evaluation of water quality The present sensor could easily discriminate several kinds of drinks such as coffee, beer and ionic drinks w16x. Fig. 6 shows the time course of responses of six channels to a commercial milk. The response electric potential is the difference between the electric potential for the measured milk and that for another commercial milk as a standard origin. Data sampling was made every second. The response was so fast that it occurred within 1 s, as soon as the electrode was immersed in the sample, and then the response curve of each channel was almost flat for the measuring time 30 s. All the channels showed very slow gradual changes of response electric potentials; this indicates the slow adsorption of chemical components of milk onto the membrane Žor desorption from the membrane.. Although this slow component of the response occurred, most of the response occurred at the initial time as soon as the membranes were immersed in the sample. A similar situation occurs for other foodstuffs such as beer, coffee and mineral water using preconditioned membranes.
Fig. 6. Time course of response electric potentials of channels 1, 2, 3, 5, 7, and 8 in the measurement of milk.
Fig. 7. Taste map of beer w7x. Symbols enclosing the brand of beer, such as dotted ellipses and squares, represent beer produced by different companies in Japan, and thin solid ellipses represent beer from other countries.
Fig. 7 shows the result of PCA applied to the response patterns for 33 brands of beer w7x. Comparison with the human taste sense indicated that PC1 corresponds to ‘‘rich taste’’ and ‘‘light taste’’, PC2 ‘‘sharp taste’’ and ‘‘mild taste’’. One of the largest merits of the taste sensor is the fact that any pretreatment of foods is unnecessary. The taste can be measured soon after drink is poured into a cup. In other words, the change in the taste with time can also be measured using the taste sensor. The sensor can also be used for analysis of the taste of gelatiniform or solid foods. When eating food, humans first masticate the food with their teeth and then taste it. Therefore, we used a mixer in place of teeth and crushed tomatoes before measuring them. As a result, different brands of tomatoes were distinguished by the shapes of the output electric potential patterns w17x. For quantification of the taste of tomatoes, the taste sensor was first applied to canned tomato juice, to which four basic taste substances such as NaCl, citric acid, MSG and glucose had been added. Data were analyzed by means of PCA. The taste of several brands of tomatoes was expressed in terms of four basic taste qualities by projecting the data obtained from these tomatoes onto the principal axes. The result agreed well with that of human taste sense. Recently, interest in the quality of drinking water has grown rapidly. Many people are requesting safe and delicious water. The importance of water has also been recognized as a result of interest in environmental pollution issues. However, no convenient measurement system appropriate for evaluation of drinking water quality is available. Fig. 8 shows the result of PCA applied to response patterns for 41 kinds of commercial mineral water w18x.
K. Toko r Sensors and Actuators B 64 (2000) 205–215
Fig. 8. Taste map of 41 kinds of mineral water w18x. The figure implies hardness of water.
The right-lower plane contains mineral water with strong hardness, i.e., hard water, whereas the left-lower plane is occupied by soft water. The taste of mineral water is quite subtle, and hence it is difficult for humans to discriminate between different brands. However, the taste sensor responded well to the different kinds of mineral water tested. In spite of the low concentrations of taste substances in mineral water, the sensor can discriminate between different brands because of its high sensitivity to ions. The present result reveals a possible application of the taste sensor to environmental measurements of the quality of water. Detection of some toxic substances in factory drains is a time-consuming process because many kinds of substances need to be analyzed. However, it is quite necessary to check quickly the safety of water before drinking in daily life. The taste sensor is the most adequate for this purpose, because it can respond simultaneously to many kinds of chemical substances with high sensitivity. We can judge whether drinking water is safe or not only if we know the safety range of response electric pattern of the sensor output. The taste sensor was applied to measurement of contamination of factory drains w19,20x. Many pollutants such as CNy, Fe 3q and Cu2q could be measured in a few minutes with the detection limits lower than regulations of drain. Cyanide was detected selectively using multiple regression analysis.
211
It is well-known that bitterness is suppressed by coexistence of sweet substances such as sucrose. To quantify the bitterness, it is necessary to measure the taste by taking account of the suppression effect. We tried to detect this effect using the taste sensor w21x. The measurement was made on the samples composed of 1 mM quinine and coexistent sucrose with different concentrations. As a result, the response pattern shifted downward as a whole with increasing sucrose concentration. The decrease in response potentials of channels 1–5 was as large as 10 mV or more, and is opposite to the change in response potentials for increasing quinine concentration. Therefore, it may be possible to regard this decrease as the equivalent decrease in quinine concentration Žor decrease in bitterness. with increasing sucrose concentration. We applied the PCA to the data on quinine in order to quantify bitterness. Fig. 9 shows the relationship between the PC1 and the quinine concentration w21x. The contribution rates of the original data to PC1, PC2 and PC3 were 95.4%, 3.4%, and 0.7%, respectively. This means that we can discuss the data on quinine using only PC1, which is expressed by the formula: PC1 s a1 Ž Õ 1 y ÕX1 . q a 2 Ž Õ 2 y ÕX2 . q PPP qa7 Ž Õ 7 y ÕX7 . , Ž 2. where a i is the factor loading, Õi the response potential of channel i, and ÕXi is the response potential of channel i averaged over the measured samples. In the present case, a i and ÕiX were determined as a i s Ž0.477, 0.324, 0.590, 0.358, 0.172, y0.317, y0.249. and ÕXi s Ž146.4, 72.39, 43.81, 47.57, 55.12, 79.70, 52.54.. In Fig. 9, a straight line is drawn using the method of least squares. It can be seen that PC1 increases in proportion to the quinine concentration in logarithmic scale. It is noticeable that this relation agrees with the well-known Weber–Fechner’s law of human sensing evaluation. This
5. Suppression of bitterness It is important especially for pharmaceutical and food sciences to express the extent of bitterness, e.g., in the case of developing syrups. To date, however, the main method of measurement is the sensory evaluation by humans, in which tasting bitterness stresses to inspectors, and conventional chemical analyses are the subsidiary methods. Therefore, taste-sensing devices, which can detect the bitterness, have been desired for a long time.
Fig. 9. The relationship between PC1 data and the quinine concentration. The straight line was obtained by the method of least squares w21x.
212
K. Toko r Sensors and Actuators B 64 (2000) 205–215
Fig. 10. The suppression of bitterness of quinine by sucrose expressed by the t scale w21x.
law states that a sensation is proportional to the logarithm of stimulus intensity. The result of Fig. 9 means that PC1 can be considered to express the strength of bitterness. Therefore, we can transform the PC1 value obtained from the sensor output to the t scale w22x as the measure of bitterness to express the reported relationship between the t scale and quinine concentration:
t s 2.35 log Ž cr0.00011 . ,
Ž 3.
where c is the quinine concentration Žgr100 cm3 .. The least squares line in Fig. 9 is given by: PC1 s 114.75 log C q 1.05
Ž 4.
with C denoting the molar concentration; hence, we get a relationship between t and PC1:
t s 0.0205PC1 q 5.98.
little weaker suppression effect was found using only PA, which may have a structure similar to that of a liposome in aqueous solution. Addition of 1% PA to the solutions containing bitter substances such as quinine and propranolol decreased the bitterness to the extent that humans sense no or a very weak bitter taste. This suppression effect was studied w24,25x using the taste sensor and a commercial bitter-masking substance ŽBMI-60. composed of phospholipids; its ingredients are 15–20% PA, 5% phosphatidylcholine, 40% phosphatidylinositol and 10–15% phosphatidylethanolamine. The measurement and data processing methods were the same as those described above. Fig. 11 shows the change in equivalent quinine concentration with increasing concentration of the above phospholipids. The equivalent quinine concentration implies the concentration estimated from PC1 in Eq. 4, and hence directly expresses the bitter strength felt by humans. Two quinine concentrations of 0.1 and 1 mM were studied, and two methods of application of the phospholipids to the quinine solution were tried. One method is the measurement of the mixed solution of quinine and the phospholipids, and another is the measurement of quinine solution using an electrode that had been pretreated by immersing it in a solution containing the phospholipids for 10 s. The former method permits the possibility that the phospholipids directly interact with quinine, while the latter contains the modification of the lipidrpolymer membranes owing to the presumable adsorption of the phospholipids. Using a mixed solution containing 0.1 mM quinine, addition of 0.1% phospholipids decreases the equivalent quinine concentration to about 8 mM, which elicits no bitter taste. For the solution of 1 mM quinine, 0.7% and 1% addition resulted in a weak or absent bitterness. In the second method where the quinine solution is measured using the lipidrpolymer membranes modified by adsorption of the phospholipids, the bitterness of 0.1 mM quinine
Ž 5.
If the PC1 value is calculated, we can estimate the strength of bitterness expressed by the t scale. The strength of bitterness was then obtained from the response potentials for samples, which contain sucrose, using Eqs. 2 and 5. The result is shown as a function of the sucrose concentration in Fig. 10. We can see a decrease of the strength of bitterness with increasing sucrose concentration despite a constant quinine concentration 1 mM. This result implies a satisfactory detection of the suppression of bitterness induced by sucrose. The same measurement and procedure as the above was made on a drug substance. As a result, the decrease of the bitter strength of a drug substance with increasing sucrose concentration was quantified. A lipoprotein ŽPA–LG. made of phosphatidic acid ŽPA. and b-lactoglobulin ŽLG. completely suppressed bitterness without affecting other taste qualities w23x. A similar but a
Fig. 11. Changes of the equivalent quinine concentration from 1 mM Ža. and 0.1 mM Žb. quinine as a result of addition of phospholipids. The result using a solution of mixed quinine and phospholipids is denoted with a dashed line, while that for quinine solution using the phospholipid-modified electrode is denoted by the solid line.
K. Toko r Sensors and Actuators B 64 (2000) 205–215
was effectively weakened to the zero level beyond 0.3% phospholipids. The bitterness of 1 mM quinine was also mostly removed. The present method using the taste sensor can be expected to provide a new automated method to measure the strength of the bitterness of a drug substance to replace human sensory evaluation. In addition, study of the taste interaction using the taste sensor will contribute to clarification of reception mechanisms in the gustatory system.
6. Taste sensing FET (TSFET) Depending on the purpose and objective to be measured, it may be necessary to miniaturize the taste sensor. In fact, the taste sensor will be able to be applied to real-time measurements of chemical substances in organism as well as foodstuffs. The present work is concerned with the first step of development of an integrated taste sensor w26x. Fig. 12 shows the detecting part made of MOSFET, on the gate of which the lipidrpolymer membrane is pasted. The size of FET is 1 mm width= 2 mm length= 0.5 mm thickness, and the gate part has the open 10 = 340 mm2 area. The membrane was made in the same way as above. The membrane of 1 = 2 mm size was pasted on the gate using 1% PVCrTHF solution, which inhibited easy detachment of the membrane. Eight MOSFET electrodes with the different lipidrpolymer membrane were separately prepared, and the gate–source voltage was measured by keeping the source–drain current constant. The output was taken into a computer using a handmade scanner for exchanging sequentially the eight outputs. Four kinds of foodstuffs as coffee, canned coffee, beer and sport drink were measured; these foodstuffs were discriminated well because of very small standard deviations, e.g., 0.82, 0.47, 0.47 and 0.83 mV in DOP, OAm, TOMA and 5:5 membranes, respectively, for sport drink. The FET taste sensor was found to show the same characteristics as the conventional taste sensor. It is due to the utilization of the same lipidrpolymer membranes, i.e.,
Fig. 12. Detecting part with lipidrpolymer membrane w26x.
213
the output Ž Vgs . is the change in membrane potential brought about by interactions between the membrane and chemical substances. However, the stability was not so good as the conventional taste sensor, i.e., the lipidrpolymer membranes were detached from the MOSFET after use of over 100 h, although the 1% PVCrTHF treatment was very effective to prolong the lifetime. If this treatment was not made, the stable response potential was not obtained sometimes from the first time. A dihexadecyl phosphate ŽDHP. Langmuir–Blodgett membrane was also used. The PVC membrane was first formed on the FET surface using a casting method to make hydrophobic chains of DHP being stably attached to the gate part. As a result, responses to five chemical substances such as quinine and HCl were different from the responses of lipidrpolymer membranes as above. In the present study, a real fabrication of integrated taste sensor was not made. It will be possible based on the present results. These trials will open doors to fabrication of an integrated taste sensing system.
7. Detection of wine flavor using taste sensor and electronic nose Wine has both taste and odor qualities due to different aromatic molecules in the liquid and vapor phases. The average wine contains about 80–85% water and over 500 different substances, some of which are very important to the flavor of wine in spite of their low concentrations. Wine is a suitable candidate for testing the performance of the sensory fusion of taste sensor and odor sensor Ži.e., electronic nose.. We studied the wine flavor using both taste sensor and electronic nose w27x. The electronic nose array is composed of four different conducting polymers. The monomer Ž25 mg. is dissolved in trichloroethylene Ž2 ml. and the oxidizing salt previously dissolved in acetonitrile is added in a dropwise manner. The polymerization process then occurs and the resulting solution is sprayed onto an alumina substrate where four interdigitated electrodes were previously evaporated. After evaporating the solvent, the conducting polymer is connected with the four electrodes. Four different sensing elements were obtained by combining two different monomers and two oxidizing salts. The electric resistance measured at the inner electrodes varies when volatile molecules are adsorbed at the surface of the polymer film. The average sensitivity expressed as the ratio of the resistance change to the base resistance value was almost always less than 2% in the case of the elements used in this work for wine sensing. Eight- and four-dimensional data arrays were obtained from measurements using taste sensor and electronic nose, respectively. A normalization procedure is necessary, because the output quantity is different between these two kinds of sensors.
214
K. Toko r Sensors and Actuators B 64 (2000) 205–215
8. Summary
Fig. 13. Results of the PCA applied to the combination of the data set from the taste sensor with the data set from the electronic nose w27x. Wine 1 Žwhite.: Est! Est! Est! di Montefiascone 1995, Italy; wine 2 Žred.: Bon Marche’ Mercian, Japan; wine 4 Žwhite.: Chablis 1994, France; and wine 5 Žred.: Rosso di Montalcino, Fattoria dei Barbi 1994, Italy.
After either set of data was normalized, a 12-dimensional data array was obtained for each measurement on four different wines. The combination of the two sets of data has led to a new representation of the samples in the 12-dimensional space, which simultaneously contains information from taste sensor and electronic nose concerning the sample measured. The PCA was performed on this set of data and the results are shown in Fig. 13. The discrimination among four wines is satisfactory. The relative positioning of the clusters in the principal component plane was similar to the case of the electronic nose and the relatively high distance between clusters of the red Ž2 and 5. and white Ž1 and 4. wines were successfully achieved by the contribution of the taste sensor. The effect of the aging process after opening on the quality of wine was also studied. It was found that the system can discriminate among differently aged samples of the same red wine. The sensor fusion is very effective because the information provided by the different arrays are, to some extent, independent from each other; they account for different characteristics of the wines themselves. Conventional multiple sensor arrays have several sensing elements belonging to the same kind of technology, e.g., conducting polymer sensor array, metal oxide sensor array and lipid membrane sensor array. These arrays show broad sensitivities to certain groups of substances, but on the other hand, they are not sensitive to other compounds. If different kinds of sensor technologies are simultaneously applied, provided that the information from the different sources is independent, it is worthwhile to combine them to obtain a broader viewpoint of the samples measured.
The taste sensor has a new concept of global selectivity. What is important in recognition of taste is not discrimination of minute difference in molecular structures but to transform molecular information contained in interactions with biological membranes into several kinds of groups, i.e., taste intensities and qualities. This is high-level function, where intelligent sensing is required. In this meaning, the taste sensor is essentially an intelligent sensor to reproduce the taste sense, which is a complex, comprehensive sense of humans. Combination of different sensors such as taste sensor and electronic nose may drastically increase obtained information. Wine flavour was successfully discriminated using the taste sensor and the electronic nose that utilizes conducting polymers. The sense of taste largely depends on subjective factors of human feelings. If we compare the standard index measured by means of the taste sensor with the sensory evaluation, we will be able to assess taste objectively. Moreover, this study will contribute to clarification of the mechanism of information processing of taste in the brain as well as the reception at taste cells.
References w1x C. Pfaffmann, in: Handbook of Physiology, American Physiological Society, Washington, DC, 1959, pp. 507–533. w2x Y. Ninomiya, M. Funakoshi, Qualitative discrimination among umami and the four basic taste substances in mice, in: Y. Kawamura, M.R. Kare ŽEds.., Umami: A Basic Taste, Marcel Dekker, New York, 1987, pp. 365–385. w3x K. Hayashi, M. Yamanaka, K. Toko, K. Yamafuji, Multichannel taste sensor using lipid membranes, Sens. Actuators B 2 Ž1990. 205–213. w4x K. Toko, K. Hayashi, M. Yamanaka, K. Yamafuji, Multi-channel taste sensor with lipid membranes, in: Tech. Dig. 9th Sens. Symp., Tokyo, Japan,1990, pp. 193–196. w5x H. Ikezaki, K. Hayashi, T. Yamanaka, R. Tatsukawa, K. Toko, K. Yamafuji, Multichannel taste sensor using lipid membranes, Trans. IEICE Jpn. J74-C-II Ž1991. 434–442, Žin Japanese.. w6x Y. Kikkawa, K. Toko, T. Matsuno, K. Yamafuji, Discrimination of taste of amino acids with a multichannel taste sensor, Jpn. J. Appl. Phys. 33 Ž1994. 5731–5736. w7x K. Toko, A taste sensor, Meas. Sci. Technol. 9 Ž1998. 1919–1936. w8x K. Oohira, K. Toko, H. Akiyama, H. Yoshihara, K. Yamafuji, Electric characteristics of hybrid polymer membranes composed of two lipid species, J. Phys. Soc. Jpn. 64 Ž1995. 3554–3561. w9x K. Oohira, K. Toko, Theory of electric characteristics of lipidrPVCrDOPP membrane and PVCrDOPP membrane in response to taste stimuli, Biophys. Chem. 61 Ž1996. 29–35. w10x K. Toko, T. Fukusaka, Measurement of hydrophobicity of amino acids using a multichannel taste sensor, Sens. Mater. 9 Ž1997. 171–176. w11x T. Nagamori, K. Toko, Quantification of taste of amino acids with a taste sensor, in: Dig. Tech. Papers of Transducers ’99, Sendai, Japan,1999, pp. 62–65.
K. Toko r Sensors and Actuators B 64 (2000) 205–215 w12x T. Ninomiya, S. Ikeda, S. Yamaguchi, T. Yoshikawa, Study of taste of amino acids, in: Proc. 7th Congress Sensory Tests, Tokyo, Japan,1966, pp. 109–123, Žin Japanese.. w13x G.G. Birch, Structure, chirality and solution properties of glutamates in relation to taste, in: Y. Kawamura, M.R. Kare ŽEds.., Umami: A Basic Taste, Marcel Dekker, New York, 1987, pp. 173–184. w14x J. Kirimura, A. Shimizu, A. Kimizuka, T. Ninomiya, N. Katsuya, The contribution of peptides and amino acids to the taste of foodstuffs, J. Agric. Food Chem. 17 Ž1969. 689–692. w15x K. Toko, T. Nagamori, Quantitative expression of mixed taste of amino acids using multichannel taste sensor, Trans. IEE Jpn. 119-E Ž1999. 528–531. w16x K. Toko, T. Matsuno, K. Yamafuji, K. Hayashi, H. Ikezaki, K. Sato, S. Kawarai, Multichannel taste sensor using electric potential changes in lipid membranes, Biosens. Bioelectron. 9 Ž1994. 359–364. w17x Y. Kikkawa, K. Toko, K. Yamafuji, Taste sensing of tomatoes with a multichannel taste sensor, Sens. Mater. 5 Ž1993. 83–90. w18x K. Toko, Electronic sensing of tastes, in: H. Baltes, W. Gopel, J. ¨ Hesse ŽEds.., Sensors Update 3 Wiley–VCH, Weinheim, 1998, pp. 131–160. w19x A. Taniguchi, Y. Naito, N. Maeda, H. Ikezaki, K. Toko, Multichannel lipid membrane sensor for evaluation of water quality, Trans. IEE Jpn. 118-E Ž1998. 634–640, Žin Japanese.. w20x A. Taniguchi, Y. Naito, N. Maeda, Y. Sato, H. Ikezaki, Development of monitoring system of water quality using taste sensor, Sens. Mater. 11 Ž1999. 437–446. w21x S. Takagi, K. Toko, K. Wada, H. Yamada, K. Toyoshima, Detection
w22x w23x w24x
w25x w26x w27x
215
of suppression of bitterness by sweet substance using a multichannel taste sensor, J. Pharm. Sci. 87 Ž1998. 552–555. T. Indow, An application of the t scale of taste, Percept. Psychophysiol. 5 Ž1969. 347–351. Y. Katsuragi, K. Kurihara, Specific inhibitor for bitter taste, Nature 365 Ž1993. 213–214. S. Takagi, K. Toko, K. Wada, T. Ohki, Detection of suppression of bitterness by phospholipids using electronic tongue, in: Dig. Tech. Papers of Transducers ’99, Sendai, Japan,1999, pp. 1638–1641. K. Toko, Biominetic Sensor Technology, Cambridge Univ. Press, Cambridge, 2000, Chap. 6, in press. K. Toko, R. Yasuda, S. Ezaki, T. Fujiyoshi, Taste sensing FET ŽTSFET., Trans. IEE Jpn. 118-E Ž1998. 1–5. S. Baldacci, T. Matsuno, K. Toko, R. Stella, D. De Rossi, Discrimination of wine using taste and smell sensors, Sens. Mater. 10 Ž1998. 185–200.
Biography Kiyoshi Toko received his B.E., M.E. and Ph.D. degrees, all in electrical engineering, from Kyushu University in 1975, 1977, and 1982, respectively. He is now a Professor of the Graduate School of Information Science and Electrical Engineering of Kyushu University. Dr. Toko is a member of the Japan Society of Applied Physics, the Institute of Electrical Engineers of Japan and the Japanese Association for the Study of Taste and Smell.
Taste sensor Kiyoshi Toko Department of Electronic DeÕice Engineering, Graduate School of Information Science and Electrical Engineering, Kyushu UniÕersity, Fukuoka, 812-8581, Japan
Abstract A taste sensor with global selectivity, i.e., electronic tongue, is composed of several kinds of lipidrpolymer membranes for transforming information of taste substances into electric signal. The sensor output shows different patterns for chemical substances which have different taste qualities such as saltiness and sourness. Taste interactions such as suppression effect, which occurs between bitterness and sweetness, can be detected and quantified using the taste sensor. Amino acids and peptides can be classified into several groups according to their own tastes from sensor outputs. Bitter-tasting amino acids such as L-tryptophan have response electric patterns similar to a typical bitter substance, quinine. The taste of foodstuffs such as beer, sake, coffee, mineral water, milk and vegetables can be discussed quantitatively. The taste sensor with lipid membranes provides the objective scale for the human sensory expression and will contribute to clarification of the reception mechanism at gustatory cells. q 2000 Elsevier Science S.A. All rights reserved. Keywords: Taste sensor; Chemical substances; Global selectivity; Quantified taste; Amino acids; Lipid membranes
1. Introduction Taste is comprised of five basic qualities w1,2x: sourness produced by hydrogen ions of HCl, acetic acid, citric acid, etc.; saltiness produced mainly by NaCl; sweetness due to sucrose, glucose, etc.; bitterness produced by quinine, caffeine and MgCl 2 . The last is umami taste, which is the Japanese term for implying ‘‘deliciousness,’’ produced by monosodium glutamate ŽMSG. contained in seaweeds, disodium inosinate ŽIMP. in meat and fish and disodium guanylate ŽGMP. in mushrooms. In a gustatory system, substances producing taste are received by the biological membrane of gustatory cells in taste buds on tongue. Information on taste substances is transduced into an electric signal, which is transmitted along the nerve fiber to the brain, where the taste is perceived. We have developed a taste-sensing system whose transducer is composed of lipidrpolymer membranes w3–5x. The output of this system is not the amount of specific taste substances but the taste quality and intensity, because different output electric patterns are obtained for chemical substances producing different taste qualities such as sourness and saltiness. On the other hand, similar patterns are obtained for chemical substances producing the same taste, such as MSG, IMP and GMP, which have an umami taste, and NaCl, KCl and KBr for saltiness. The development of this sensor is based on a concept very different from that of conventional chemical sensors,
which selectively detect specific chemical substances such as glucose or urea. However, taste cannot be measured even if all the chemical substances contained in foodstuffs are measured. Humans do not distinguish each chemical substance, but express the taste itself; the relationship between chemical substances and taste is not clear. It is also not practical to arrange so many chemical sensors as chemical substances, which number over 1000 in one kind of foodstuff. Moreover, there exist interactions between taste substances, such as a suppression effect. Sweet substances suppress the taste intensity of bitter substances. Discrimination of each chemical substance is not important here, but recognition of the taste itself and its quantitative expression must be made. The taste sensor using lipidrpolymer membranes has a concept of global selectivity, which implies the ability to classify enormous kinds of chemical substances into several groups.
2. Taste sensor 2.1. Multichannel electrode Eight kinds of lipids were used for preparing the membranes; used lipids were, e.g., oleic acid ŽOA., oleyl amine ŽOAm. and decyl alcohol ŽDA. as listed in Table 1. Depending upon the object to be measured, we prepared
0925-4005r00r$ - see front matter q 2000 Elsevier Science S.A. All rights reserved. PII: S 0 9 2 5 - 4 0 0 5 Ž 9 9 . 0 0 5 0 8 - 0
K. Toko r Sensors and Actuators B 64 (2000) 205–215
206
Table 1 Lipid materials used in the multichannel electrode Channel
Lipid Žabbreviation.
1 2 3
n-Decyl alcohol ŽDA. Oleic acid ŽOA. Dioctyl phosphate ŽBisŽ2-ethylhexyl.hydrogen phosphate, DOP. DOP: TOMA s9:1 DOP: TOMA s 5:5 DOP: TOMA s 3:7 Trioctyl methyl ammonium chloride ŽTOMA. Oleyl amine ŽOAm.
4 5 6 7 8
different lipid materials. For example, the mixed, hybrid membranes composed of dioctyl phosphate ŽDOP. and trioctyl methyl ammonium chloride ŽTOMA. were used for measurements of amino acids w6x. The lipidrpolymer membrane was a transparent, soft film with approximately 200 mm thickness. Each lipidrpolymer membrane was fitted on the part of a plastic tube, which has a hole, such that the inner part of the cylinder is isolated from the outside. The end of the cylinder was sealed with a stopper that holds an AgrAgCl electrode. The tube was filled with 3 M KCl solution. Eight detecting electrodes thus prepared were separated into two groups, and connected to two electrode holders, which are controlled mechanically by a robot arm, as illustrated in Fig. 1. 2.2. Responses to fiÕe primary tastes Fig. 2 shows the electric potential pattern from seven channels for two kinds of taste qualities of sourness and
saltiness, measured by taking the origin to 1 mM KCl w7x. The patterns of substances producing different taste qualities are much different, and hence, each taste can be easily discriminated. The reproducibility was very high, because the standard deviations were smaller than 1%. On the other hand, the taste sensor has similar response patterns to the same group of taste, i.e., as examples of sour substances, HCl, citric acid and acetic acid show similar response patterns. Salty substances, NaCl, KCl and KBr show similar patterns, too. The same result also holds for other taste qualities such as bitterness, sweetness and umami taste. The taste sensor can respond to the taste itself. This fact implies that the taste sensor has global selectivity. Of course, the taste sensor has an ability of molecular recognition to distinguish chemical substances even in the same group of taste, e.g., the patterns for HCl, citric acid and acetic acid are slightly different. The reception mechanism in lipidrpolymer membranes of the taste sensor was clarified quantitatively based on electrochemical theory, which treats the surface electric potential, surface charge density and binding of ions such as protons and hydrophobic ions w8,9x. This result suggests participation of a similar mechanism at the gustatory reception level.
3. Amino acids Each channel responded to amino acids in different ways depending on their tastes w6x. L-Tryptophan, which elicits almost pure bitter taste in humans, increased the potentials of channels 1, 2 and 3 greatly. This tendency was observed for other amino acids, which mainly exhibit
Fig. 1. Taste-sensing system ŽSA402, Anritsu..
K. Toko r Sensors and Actuators B 64 (2000) 205–215
207
Fig. 2. Response electric potential patterns for sour and salty substances w7x.
bitter taste, such as L-phenylalanine and L-isoleucine. On the other hand, L-alanine, glycine and L-threonine taste mainly sweet. For these amino acids, the potentials of channels 1 and 2 decreased. L-Valine and L-methionine, which taste bitter and slightly sweet, decreased the potential of channel 5; the responses of channels 1 and 2 were small. Fig. 3 shows a diagram obtained by the principal component analysis ŽPCA. w6x. The first principal axis ŽPC1. reflects bitterness and sweetness, while the second principal axis ŽPC2. reflects sourness and umami taste. Amino acids are roughly classified into five groups. The taste changes gradually from sweetness for glycine to bitterness for L-tryptophan along the PC1 axis. In addition, a strong correlation between the output of the taste sensor and hydrophobicity of amino acids was found w10x. Next, we compared the response patterns for bitter-tasting amino acids such as L-tryptophan with that of quinine, which is a typical bitter substance. Of course, these two chemicals have very different chemical structures; nevertheless, we feel the same bitter taste quality. Fig. 4 shows the comparison of response patterns of three amino acids ŽL-alanine, L-tryptophan, L-phenylalanine. with those of a bitter substance Žquinine., sour substance ŽHCl. and umami taste substance ŽMSG. w11x. The response patterns were normalized using the formula: Õi s
Vi
(
,
8
Ý < Vi <
Fig. 4 includes three normalized patterns for one chemical substance with three different concentrations. By the normalization procedure of Eq. 1, the three patterns of each chemical substance agree with each other; that is, the pattern is independent of the concentration. This fact implies that each chemical substance has an original pattern characteristic of each taste quality. The patterns of amino acids such as L-tryptophan and L-phenylalanine are similar to that of quinine. However, they differ from those of other taste substances such as L-alanine, HCl and MSG. The patterns for L-tryptophan and L-phenylalanine jut out to the upper-right direction Ži.e., at channels 1–3., whereas the pattern for L-alanine shows the opposite tendency: its pattern bulges out to the lower-left direction. The bitter chemical quinine shows a bulge to the upper-right direction in a similar way to L-tryptophan.
Ž 1.
2
is1
where Vi denotes the response electric potential of channel i. To make it easy to see the pattern, the response electric potentials of the positively charged membranes Žchannels 6–8. in Table 1 and the non-charged membrane Žchannel 5. were reversed, because they usually have a sign opposite to those of the negatively charged membranes Žchannels 1–4..
Fig. 3. Discrimination of taste of amino acids using the taste sensor w6x.
208
K. Toko r Sensors and Actuators B 64 (2000) 205–215
Fig. 4. Normalized patterns for L-tryptophan Ža., L-phenylalanine Žb., quinine Žc., L-alanine Žd., HCl Že., and MSG Žf. w11x. The three numerical figures closed by a square attached to each pattern imply the concentration ŽmM..
The correlation coefficients of the pattern between Ltryptophan Ž10 mM. and five basic taste substances are as follows: 0.903 for 0.3 mM quinine, 0.276 for 30 mM NaCl, 0.763 for 3 mM HCl, 0.515 for 100 mM sucrose and 0.408 for 10 mM MSG. Although L-tryptophan has low correlations with salty ŽNaCl., sweet Žsucrose. and umami ŽMSG. taste substances, it has a high correlation with a bitter substance, quinine. This result indicates that L-tryptophan shows the same taste as quinine, i.e., L-tryptophan tastes bitter.
Comparison of the original response pattern of L-tryptophan with that of quinine makes it possible to estimate the bitter strength of L-tryptophan in terms of the quinine concentration Žsee also Fig. 9.. As a result, it was concluded that 10 mM L-tryptophan has the same bitter strength as 0.02 mM quinine. To confirm it, we performed the sensory evaluation using our tongue. The result supported the estimation using the taste sensor: humans felt the same bitter strength as 10 mM L-tryptophan by taking 0.02–0.03 mM quinine.
K. Toko r Sensors and Actuators B 64 (2000) 205–215
Fig. 5. Normalized patterns for L-methionine Ža. and the mixed solution Žb., which contains 100 mM L-alanine and L-tryptophan with three different concentrations Ž1, 3, 10 mM. w11x.
This result implies that the taste sensor measures the taste in itself, as humans do, irrespective of the difference of chemical structures of amino acids and alkaloids such as quinine. L-Methionine and L-valine shows bitter and sweet tastes simultaneously w12–14x. We tried to produce the complicated mixed taste of these amino acids using the combina-
209
tion of a sweet amino acid, L-alanine, and a bitter amino acid, L-tryptophan. Fig. 5 shows the normalized patterns for L-methionine and the mixed solution, which contains 100 mM L-alanine and L-tryptophan at 1, 3 and 10 mM w11x. Three patterns for L-methionine at three different concentrations agree fairly well with each other. In addition, they differ from those for either L-alanine or L-tryptophan alone, as shown in Fig. 4. For example, 10 mM L-methionine has the low correlations of 0.22 and 0.57 with 10 mM L-alanine and 1 mM L-tryptophan, respectively. This implies that Lmethionine has an original taste quality that is different from pure sweet or bitter taste. The patterns for the mixed solution composed of 100 mM L-alanine and L-tryptophan of three different concentrations do not agree with each other. Increasing the Ltryptophan concentration tends to cause a bulge to the upper-right direction. Since the pattern for L-tryptophan alone has a characteristic bulge of this type, it implies that the taste becomes nearer that of L-tryptophan, as expected. Table 2 summarizes the correlation coefficients among L-methionine, L-alanine, L-tryptophan and the above three mixed solutions w15x. The response patterns for L-alanine have negative correlations with those for L-tryptophan. It agrees with the fact that they show very different taste qualities. We can see that the highest correlation is realized between one of the mixed solutions and L-methionine. Increasing L-tryptophan concentration in the mixed solution leads to both the decrease of correlation with L-alanine and the increase of correlation with L-tryptophan, as considered reasonably. The highest correlation 0.99 is realized for 30 mM L-methionine using the mixed solution 100 mM L-alanine and 10 mM L-tryptophan. The sensory tests by humans agreed with this result. A similar result was obtained for L-valine, which elicits bitter and sweet tastes simultaneously. The sensor output showed the highest correlation between 30 mM L-valine and 100 mM L-alanine plus 3 mM L-tryptophan, as felt by humans.
Table 2 Correlation coefficients among L-methionine, L-alanine, L-tryptophan and the three mixed solutions w15x. The unit of the concentrations is mM; e.g., for L-methionine, 10, 30, and 100 mM were chosen. In the case of mixed solutions composed of 100 mM L-alanine and L-tryptophan, the L-tryptophan concentration ŽmM. is shown Methionine 10 Methionine
Mixed
Alanine
Tryptophan
10 30 100 1 3 10 10 30 100 1 3 10
0.96 0.93 0.93 0.96 0.93 0.22 0.26 0.34 0.57 0.60 0.61
Mixed
Alanine
30
100
1
3
10
0.96
0.93 0.99
0.93 0.90 0.89
0.96 0.97 0.96 0.95
0.93 0.99 0.98 0.88 0.98
0.99 0.90 0.97 0.99 0.10 0.13 0.23 0.66 0.70 0.73
0.89 0.96 0.98 0.11 0.14 0.24 0.64 0.68 0.72
0.95 0.88 0.48 0.52 0.60 0.33 0.38 0.40
0.98 0.20 0.24 0.34 0.59 0.64 0.65
0.02 0.06 0.16 0.72 0.76 0.78
10 0.22 0.10 0.11 0.48 0.20 0.02 1.00 0.99 y0.66 y0.62 y0.61
Tryptophan 30 0.26 0.13 0.14 0.52 0.24 0.06 1.00 0.99 y0.63 y0.59 y0.58
100 0.34 0.23 0.24 0.60 0.34 0.16 0.99 0.99 y0.56 y0.51 y0.49
1
3
10
0.57 0.66 0.64 0.33 0.59 0.72 y0.66 y0.63 y0.56
0.60 0.70 0.68 0.38 0.64 0.76 y0.62 y0.59 y0.51 1.00
0.61 0.73 0.72 0.40 0.65 0.78 y0.61 y0.58 y0.49 0.99 1.00
1.00 0.99
1.00
210
K. Toko r Sensors and Actuators B 64 (2000) 205–215
Dipeptides elicit various taste qualities according to the combination of two amino acids. For bitter dipeptides such as Gly–Leu, Gly–Phe and Leu–Gly, the taste sensor showed similar response patterns for L-tryptophan. In addition, the patterns characteristic of sourness, which is elicited by amino acids such as L-glutamic acid and L-histidine monohydrochloride, were obtained for dipeptides, such as Gly–Asp, Ser–Glu, Ala–Glu and Gly–Gly, which taste sour. On the other hand, dipeptides such as Ala–Gly and Gly–Glu have little or no taste. Small response patterns were obtained for these dipeptides, as expected. It is believed that receptors of amino acids differ physiologically from those of alkaloids such as quinine. As found here, however, the bitterness of amino acids can be expressed using the response patterns of the taste sensor, which uses the lipid membranes, in a way similar to the bitterness of quinine. The mixed taste of amino acids can also be produced using the taste sensor. This result suggests that it may be the lipid Žhydrophobic. part of biological membranes that forms the receptor for bitter taste. 4. Quantification of taste of foods and evaluation of water quality The present sensor could easily discriminate several kinds of drinks such as coffee, beer and ionic drinks w16x. Fig. 6 shows the time course of responses of six channels to a commercial milk. The response electric potential is the difference between the electric potential for the measured milk and that for another commercial milk as a standard origin. Data sampling was made every second. The response was so fast that it occurred within 1 s, as soon as the electrode was immersed in the sample, and then the response curve of each channel was almost flat for the measuring time 30 s. All the channels showed very slow gradual changes of response electric potentials; this indicates the slow adsorption of chemical components of milk onto the membrane Žor desorption from the membrane.. Although this slow component of the response occurred, most of the response occurred at the initial time as soon as the membranes were immersed in the sample. A similar situation occurs for other foodstuffs such as beer, coffee and mineral water using preconditioned membranes.
Fig. 6. Time course of response electric potentials of channels 1, 2, 3, 5, 7, and 8 in the measurement of milk.
Fig. 7. Taste map of beer w7x. Symbols enclosing the brand of beer, such as dotted ellipses and squares, represent beer produced by different companies in Japan, and thin solid ellipses represent beer from other countries.
Fig. 7 shows the result of PCA applied to the response patterns for 33 brands of beer w7x. Comparison with the human taste sense indicated that PC1 corresponds to ‘‘rich taste’’ and ‘‘light taste’’, PC2 ‘‘sharp taste’’ and ‘‘mild taste’’. One of the largest merits of the taste sensor is the fact that any pretreatment of foods is unnecessary. The taste can be measured soon after drink is poured into a cup. In other words, the change in the taste with time can also be measured using the taste sensor. The sensor can also be used for analysis of the taste of gelatiniform or solid foods. When eating food, humans first masticate the food with their teeth and then taste it. Therefore, we used a mixer in place of teeth and crushed tomatoes before measuring them. As a result, different brands of tomatoes were distinguished by the shapes of the output electric potential patterns w17x. For quantification of the taste of tomatoes, the taste sensor was first applied to canned tomato juice, to which four basic taste substances such as NaCl, citric acid, MSG and glucose had been added. Data were analyzed by means of PCA. The taste of several brands of tomatoes was expressed in terms of four basic taste qualities by projecting the data obtained from these tomatoes onto the principal axes. The result agreed well with that of human taste sense. Recently, interest in the quality of drinking water has grown rapidly. Many people are requesting safe and delicious water. The importance of water has also been recognized as a result of interest in environmental pollution issues. However, no convenient measurement system appropriate for evaluation of drinking water quality is available. Fig. 8 shows the result of PCA applied to response patterns for 41 kinds of commercial mineral water w18x.
K. Toko r Sensors and Actuators B 64 (2000) 205–215
Fig. 8. Taste map of 41 kinds of mineral water w18x. The figure implies hardness of water.
The right-lower plane contains mineral water with strong hardness, i.e., hard water, whereas the left-lower plane is occupied by soft water. The taste of mineral water is quite subtle, and hence it is difficult for humans to discriminate between different brands. However, the taste sensor responded well to the different kinds of mineral water tested. In spite of the low concentrations of taste substances in mineral water, the sensor can discriminate between different brands because of its high sensitivity to ions. The present result reveals a possible application of the taste sensor to environmental measurements of the quality of water. Detection of some toxic substances in factory drains is a time-consuming process because many kinds of substances need to be analyzed. However, it is quite necessary to check quickly the safety of water before drinking in daily life. The taste sensor is the most adequate for this purpose, because it can respond simultaneously to many kinds of chemical substances with high sensitivity. We can judge whether drinking water is safe or not only if we know the safety range of response electric pattern of the sensor output. The taste sensor was applied to measurement of contamination of factory drains w19,20x. Many pollutants such as CNy, Fe 3q and Cu2q could be measured in a few minutes with the detection limits lower than regulations of drain. Cyanide was detected selectively using multiple regression analysis.
211
It is well-known that bitterness is suppressed by coexistence of sweet substances such as sucrose. To quantify the bitterness, it is necessary to measure the taste by taking account of the suppression effect. We tried to detect this effect using the taste sensor w21x. The measurement was made on the samples composed of 1 mM quinine and coexistent sucrose with different concentrations. As a result, the response pattern shifted downward as a whole with increasing sucrose concentration. The decrease in response potentials of channels 1–5 was as large as 10 mV or more, and is opposite to the change in response potentials for increasing quinine concentration. Therefore, it may be possible to regard this decrease as the equivalent decrease in quinine concentration Žor decrease in bitterness. with increasing sucrose concentration. We applied the PCA to the data on quinine in order to quantify bitterness. Fig. 9 shows the relationship between the PC1 and the quinine concentration w21x. The contribution rates of the original data to PC1, PC2 and PC3 were 95.4%, 3.4%, and 0.7%, respectively. This means that we can discuss the data on quinine using only PC1, which is expressed by the formula: PC1 s a1 Ž Õ 1 y ÕX1 . q a 2 Ž Õ 2 y ÕX2 . q PPP qa7 Ž Õ 7 y ÕX7 . , Ž 2. where a i is the factor loading, Õi the response potential of channel i, and ÕXi is the response potential of channel i averaged over the measured samples. In the present case, a i and ÕiX were determined as a i s Ž0.477, 0.324, 0.590, 0.358, 0.172, y0.317, y0.249. and ÕXi s Ž146.4, 72.39, 43.81, 47.57, 55.12, 79.70, 52.54.. In Fig. 9, a straight line is drawn using the method of least squares. It can be seen that PC1 increases in proportion to the quinine concentration in logarithmic scale. It is noticeable that this relation agrees with the well-known Weber–Fechner’s law of human sensing evaluation. This
5. Suppression of bitterness It is important especially for pharmaceutical and food sciences to express the extent of bitterness, e.g., in the case of developing syrups. To date, however, the main method of measurement is the sensory evaluation by humans, in which tasting bitterness stresses to inspectors, and conventional chemical analyses are the subsidiary methods. Therefore, taste-sensing devices, which can detect the bitterness, have been desired for a long time.
Fig. 9. The relationship between PC1 data and the quinine concentration. The straight line was obtained by the method of least squares w21x.
212
K. Toko r Sensors and Actuators B 64 (2000) 205–215
Fig. 10. The suppression of bitterness of quinine by sucrose expressed by the t scale w21x.
law states that a sensation is proportional to the logarithm of stimulus intensity. The result of Fig. 9 means that PC1 can be considered to express the strength of bitterness. Therefore, we can transform the PC1 value obtained from the sensor output to the t scale w22x as the measure of bitterness to express the reported relationship between the t scale and quinine concentration:
t s 2.35 log Ž cr0.00011 . ,
Ž 3.
where c is the quinine concentration Žgr100 cm3 .. The least squares line in Fig. 9 is given by: PC1 s 114.75 log C q 1.05
Ž 4.
with C denoting the molar concentration; hence, we get a relationship between t and PC1:
t s 0.0205PC1 q 5.98.
little weaker suppression effect was found using only PA, which may have a structure similar to that of a liposome in aqueous solution. Addition of 1% PA to the solutions containing bitter substances such as quinine and propranolol decreased the bitterness to the extent that humans sense no or a very weak bitter taste. This suppression effect was studied w24,25x using the taste sensor and a commercial bitter-masking substance ŽBMI-60. composed of phospholipids; its ingredients are 15–20% PA, 5% phosphatidylcholine, 40% phosphatidylinositol and 10–15% phosphatidylethanolamine. The measurement and data processing methods were the same as those described above. Fig. 11 shows the change in equivalent quinine concentration with increasing concentration of the above phospholipids. The equivalent quinine concentration implies the concentration estimated from PC1 in Eq. 4, and hence directly expresses the bitter strength felt by humans. Two quinine concentrations of 0.1 and 1 mM were studied, and two methods of application of the phospholipids to the quinine solution were tried. One method is the measurement of the mixed solution of quinine and the phospholipids, and another is the measurement of quinine solution using an electrode that had been pretreated by immersing it in a solution containing the phospholipids for 10 s. The former method permits the possibility that the phospholipids directly interact with quinine, while the latter contains the modification of the lipidrpolymer membranes owing to the presumable adsorption of the phospholipids. Using a mixed solution containing 0.1 mM quinine, addition of 0.1% phospholipids decreases the equivalent quinine concentration to about 8 mM, which elicits no bitter taste. For the solution of 1 mM quinine, 0.7% and 1% addition resulted in a weak or absent bitterness. In the second method where the quinine solution is measured using the lipidrpolymer membranes modified by adsorption of the phospholipids, the bitterness of 0.1 mM quinine
Ž 5.
If the PC1 value is calculated, we can estimate the strength of bitterness expressed by the t scale. The strength of bitterness was then obtained from the response potentials for samples, which contain sucrose, using Eqs. 2 and 5. The result is shown as a function of the sucrose concentration in Fig. 10. We can see a decrease of the strength of bitterness with increasing sucrose concentration despite a constant quinine concentration 1 mM. This result implies a satisfactory detection of the suppression of bitterness induced by sucrose. The same measurement and procedure as the above was made on a drug substance. As a result, the decrease of the bitter strength of a drug substance with increasing sucrose concentration was quantified. A lipoprotein ŽPA–LG. made of phosphatidic acid ŽPA. and b-lactoglobulin ŽLG. completely suppressed bitterness without affecting other taste qualities w23x. A similar but a
Fig. 11. Changes of the equivalent quinine concentration from 1 mM Ža. and 0.1 mM Žb. quinine as a result of addition of phospholipids. The result using a solution of mixed quinine and phospholipids is denoted with a dashed line, while that for quinine solution using the phospholipid-modified electrode is denoted by the solid line.
K. Toko r Sensors and Actuators B 64 (2000) 205–215
was effectively weakened to the zero level beyond 0.3% phospholipids. The bitterness of 1 mM quinine was also mostly removed. The present method using the taste sensor can be expected to provide a new automated method to measure the strength of the bitterness of a drug substance to replace human sensory evaluation. In addition, study of the taste interaction using the taste sensor will contribute to clarification of reception mechanisms in the gustatory system.
6. Taste sensing FET (TSFET) Depending on the purpose and objective to be measured, it may be necessary to miniaturize the taste sensor. In fact, the taste sensor will be able to be applied to real-time measurements of chemical substances in organism as well as foodstuffs. The present work is concerned with the first step of development of an integrated taste sensor w26x. Fig. 12 shows the detecting part made of MOSFET, on the gate of which the lipidrpolymer membrane is pasted. The size of FET is 1 mm width= 2 mm length= 0.5 mm thickness, and the gate part has the open 10 = 340 mm2 area. The membrane was made in the same way as above. The membrane of 1 = 2 mm size was pasted on the gate using 1% PVCrTHF solution, which inhibited easy detachment of the membrane. Eight MOSFET electrodes with the different lipidrpolymer membrane were separately prepared, and the gate–source voltage was measured by keeping the source–drain current constant. The output was taken into a computer using a handmade scanner for exchanging sequentially the eight outputs. Four kinds of foodstuffs as coffee, canned coffee, beer and sport drink were measured; these foodstuffs were discriminated well because of very small standard deviations, e.g., 0.82, 0.47, 0.47 and 0.83 mV in DOP, OAm, TOMA and 5:5 membranes, respectively, for sport drink. The FET taste sensor was found to show the same characteristics as the conventional taste sensor. It is due to the utilization of the same lipidrpolymer membranes, i.e.,
Fig. 12. Detecting part with lipidrpolymer membrane w26x.
213
the output Ž Vgs . is the change in membrane potential brought about by interactions between the membrane and chemical substances. However, the stability was not so good as the conventional taste sensor, i.e., the lipidrpolymer membranes were detached from the MOSFET after use of over 100 h, although the 1% PVCrTHF treatment was very effective to prolong the lifetime. If this treatment was not made, the stable response potential was not obtained sometimes from the first time. A dihexadecyl phosphate ŽDHP. Langmuir–Blodgett membrane was also used. The PVC membrane was first formed on the FET surface using a casting method to make hydrophobic chains of DHP being stably attached to the gate part. As a result, responses to five chemical substances such as quinine and HCl were different from the responses of lipidrpolymer membranes as above. In the present study, a real fabrication of integrated taste sensor was not made. It will be possible based on the present results. These trials will open doors to fabrication of an integrated taste sensing system.
7. Detection of wine flavor using taste sensor and electronic nose Wine has both taste and odor qualities due to different aromatic molecules in the liquid and vapor phases. The average wine contains about 80–85% water and over 500 different substances, some of which are very important to the flavor of wine in spite of their low concentrations. Wine is a suitable candidate for testing the performance of the sensory fusion of taste sensor and odor sensor Ži.e., electronic nose.. We studied the wine flavor using both taste sensor and electronic nose w27x. The electronic nose array is composed of four different conducting polymers. The monomer Ž25 mg. is dissolved in trichloroethylene Ž2 ml. and the oxidizing salt previously dissolved in acetonitrile is added in a dropwise manner. The polymerization process then occurs and the resulting solution is sprayed onto an alumina substrate where four interdigitated electrodes were previously evaporated. After evaporating the solvent, the conducting polymer is connected with the four electrodes. Four different sensing elements were obtained by combining two different monomers and two oxidizing salts. The electric resistance measured at the inner electrodes varies when volatile molecules are adsorbed at the surface of the polymer film. The average sensitivity expressed as the ratio of the resistance change to the base resistance value was almost always less than 2% in the case of the elements used in this work for wine sensing. Eight- and four-dimensional data arrays were obtained from measurements using taste sensor and electronic nose, respectively. A normalization procedure is necessary, because the output quantity is different between these two kinds of sensors.
214
K. Toko r Sensors and Actuators B 64 (2000) 205–215
8. Summary
Fig. 13. Results of the PCA applied to the combination of the data set from the taste sensor with the data set from the electronic nose w27x. Wine 1 Žwhite.: Est! Est! Est! di Montefiascone 1995, Italy; wine 2 Žred.: Bon Marche’ Mercian, Japan; wine 4 Žwhite.: Chablis 1994, France; and wine 5 Žred.: Rosso di Montalcino, Fattoria dei Barbi 1994, Italy.
After either set of data was normalized, a 12-dimensional data array was obtained for each measurement on four different wines. The combination of the two sets of data has led to a new representation of the samples in the 12-dimensional space, which simultaneously contains information from taste sensor and electronic nose concerning the sample measured. The PCA was performed on this set of data and the results are shown in Fig. 13. The discrimination among four wines is satisfactory. The relative positioning of the clusters in the principal component plane was similar to the case of the electronic nose and the relatively high distance between clusters of the red Ž2 and 5. and white Ž1 and 4. wines were successfully achieved by the contribution of the taste sensor. The effect of the aging process after opening on the quality of wine was also studied. It was found that the system can discriminate among differently aged samples of the same red wine. The sensor fusion is very effective because the information provided by the different arrays are, to some extent, independent from each other; they account for different characteristics of the wines themselves. Conventional multiple sensor arrays have several sensing elements belonging to the same kind of technology, e.g., conducting polymer sensor array, metal oxide sensor array and lipid membrane sensor array. These arrays show broad sensitivities to certain groups of substances, but on the other hand, they are not sensitive to other compounds. If different kinds of sensor technologies are simultaneously applied, provided that the information from the different sources is independent, it is worthwhile to combine them to obtain a broader viewpoint of the samples measured.
The taste sensor has a new concept of global selectivity. What is important in recognition of taste is not discrimination of minute difference in molecular structures but to transform molecular information contained in interactions with biological membranes into several kinds of groups, i.e., taste intensities and qualities. This is high-level function, where intelligent sensing is required. In this meaning, the taste sensor is essentially an intelligent sensor to reproduce the taste sense, which is a complex, comprehensive sense of humans. Combination of different sensors such as taste sensor and electronic nose may drastically increase obtained information. Wine flavour was successfully discriminated using the taste sensor and the electronic nose that utilizes conducting polymers. The sense of taste largely depends on subjective factors of human feelings. If we compare the standard index measured by means of the taste sensor with the sensory evaluation, we will be able to assess taste objectively. Moreover, this study will contribute to clarification of the mechanism of information processing of taste in the brain as well as the reception at taste cells.
References w1x C. Pfaffmann, in: Handbook of Physiology, American Physiological Society, Washington, DC, 1959, pp. 507–533. w2x Y. Ninomiya, M. Funakoshi, Qualitative discrimination among umami and the four basic taste substances in mice, in: Y. Kawamura, M.R. Kare ŽEds.., Umami: A Basic Taste, Marcel Dekker, New York, 1987, pp. 365–385. w3x K. Hayashi, M. Yamanaka, K. Toko, K. Yamafuji, Multichannel taste sensor using lipid membranes, Sens. Actuators B 2 Ž1990. 205–213. w4x K. Toko, K. Hayashi, M. Yamanaka, K. Yamafuji, Multi-channel taste sensor with lipid membranes, in: Tech. Dig. 9th Sens. Symp., Tokyo, Japan,1990, pp. 193–196. w5x H. Ikezaki, K. Hayashi, T. Yamanaka, R. Tatsukawa, K. Toko, K. Yamafuji, Multichannel taste sensor using lipid membranes, Trans. IEICE Jpn. J74-C-II Ž1991. 434–442, Žin Japanese.. w6x Y. Kikkawa, K. Toko, T. Matsuno, K. Yamafuji, Discrimination of taste of amino acids with a multichannel taste sensor, Jpn. J. Appl. Phys. 33 Ž1994. 5731–5736. w7x K. Toko, A taste sensor, Meas. Sci. Technol. 9 Ž1998. 1919–1936. w8x K. Oohira, K. Toko, H. Akiyama, H. Yoshihara, K. Yamafuji, Electric characteristics of hybrid polymer membranes composed of two lipid species, J. Phys. Soc. Jpn. 64 Ž1995. 3554–3561. w9x K. Oohira, K. Toko, Theory of electric characteristics of lipidrPVCrDOPP membrane and PVCrDOPP membrane in response to taste stimuli, Biophys. Chem. 61 Ž1996. 29–35. w10x K. Toko, T. Fukusaka, Measurement of hydrophobicity of amino acids using a multichannel taste sensor, Sens. Mater. 9 Ž1997. 171–176. w11x T. Nagamori, K. Toko, Quantification of taste of amino acids with a taste sensor, in: Dig. Tech. Papers of Transducers ’99, Sendai, Japan,1999, pp. 62–65.
K. Toko r Sensors and Actuators B 64 (2000) 205–215 w12x T. Ninomiya, S. Ikeda, S. Yamaguchi, T. Yoshikawa, Study of taste of amino acids, in: Proc. 7th Congress Sensory Tests, Tokyo, Japan,1966, pp. 109–123, Žin Japanese.. w13x G.G. Birch, Structure, chirality and solution properties of glutamates in relation to taste, in: Y. Kawamura, M.R. Kare ŽEds.., Umami: A Basic Taste, Marcel Dekker, New York, 1987, pp. 173–184. w14x J. Kirimura, A. Shimizu, A. Kimizuka, T. Ninomiya, N. Katsuya, The contribution of peptides and amino acids to the taste of foodstuffs, J. Agric. Food Chem. 17 Ž1969. 689–692. w15x K. Toko, T. Nagamori, Quantitative expression of mixed taste of amino acids using multichannel taste sensor, Trans. IEE Jpn. 119-E Ž1999. 528–531. w16x K. Toko, T. Matsuno, K. Yamafuji, K. Hayashi, H. Ikezaki, K. Sato, S. Kawarai, Multichannel taste sensor using electric potential changes in lipid membranes, Biosens. Bioelectron. 9 Ž1994. 359–364. w17x Y. Kikkawa, K. Toko, K. Yamafuji, Taste sensing of tomatoes with a multichannel taste sensor, Sens. Mater. 5 Ž1993. 83–90. w18x K. Toko, Electronic sensing of tastes, in: H. Baltes, W. Gopel, J. ¨ Hesse ŽEds.., Sensors Update 3 Wiley–VCH, Weinheim, 1998, pp. 131–160. w19x A. Taniguchi, Y. Naito, N. Maeda, H. Ikezaki, K. Toko, Multichannel lipid membrane sensor for evaluation of water quality, Trans. IEE Jpn. 118-E Ž1998. 634–640, Žin Japanese.. w20x A. Taniguchi, Y. Naito, N. Maeda, Y. Sato, H. Ikezaki, Development of monitoring system of water quality using taste sensor, Sens. Mater. 11 Ž1999. 437–446. w21x S. Takagi, K. Toko, K. Wada, H. Yamada, K. Toyoshima, Detection
w22x w23x w24x
w25x w26x w27x
215
of suppression of bitterness by sweet substance using a multichannel taste sensor, J. Pharm. Sci. 87 Ž1998. 552–555. T. Indow, An application of the t scale of taste, Percept. Psychophysiol. 5 Ž1969. 347–351. Y. Katsuragi, K. Kurihara, Specific inhibitor for bitter taste, Nature 365 Ž1993. 213–214. S. Takagi, K. Toko, K. Wada, T. Ohki, Detection of suppression of bitterness by phospholipids using electronic tongue, in: Dig. Tech. Papers of Transducers ’99, Sendai, Japan,1999, pp. 1638–1641. K. Toko, Biominetic Sensor Technology, Cambridge Univ. Press, Cambridge, 2000, Chap. 6, in press. K. Toko, R. Yasuda, S. Ezaki, T. Fujiyoshi, Taste sensing FET ŽTSFET., Trans. IEE Jpn. 118-E Ž1998. 1–5. S. Baldacci, T. Matsuno, K. Toko, R. Stella, D. De Rossi, Discrimination of wine using taste and smell sensors, Sens. Mater. 10 Ž1998. 185–200.
Biography Kiyoshi Toko received his B.E., M.E. and Ph.D. degrees, all in electrical engineering, from Kyushu University in 1975, 1977, and 1982, respectively. He is now a Professor of the Graduate School of Information Science and Electrical Engineering of Kyushu University. Dr. Toko is a member of the Japan Society of Applied Physics, the Institute of Electrical Engineers of Japan and the Japanese Association for the Study of Taste and Smell.